One of the ongoing questions these past few months has been why so many tech products have been so hard to buy. We’ve made repeated reference to known potential factors like COVID-19, economic disruptions, yield issues, and the impact of scalping bots, but there’s a new argument for what’s causing such general problems across so many markets: Insufficient investment in 200mm wafers.
Today, leading-edge silicon is invariably manufactured on 300mm wafers. Over the past few decades, manufacturers have introduced larger wafer sizes: 100mm, 150mm, 200mm, and 300mm have all been common standards at one time or another. In the PC enthusiast space, 300mm wafers have long been considered superior to 200mm wafers, because the larger wafer size reduces waste and typically improves the foundry’s output in terms of chips manufactured per day.
There aren’t that many commercial foundries still dedicated to 150mm or smaller wafer sizes, but a number of foundries still run 200mm fab lines. TSMC and Samsung both offer the node, as well as a number of second-tier foundries. GlobalFoundries has 200mm facilities, as do SMIC, UMC, TowerJazz, and SkyWater. A great many IoT and 5G chips are built on 200mm, as are some analog processors, MEMS devices, and RF solutions.
Image by Semico Research, 2018
The parts that we talk about at ExtremeTech, including GPUs, desktop and mobile CPUs, high-end cellular modems, and other similar products are only a fraction of the silicon that gets built into devices, and a lot of vendors you’ve never heard of run mature, well-understood designs through smaller, older foundry lines for minimal costs.
200mm was supposed to fade away as 300mm came online, and that worked from 2007 – 2014, but the trend reversed thereafter. Customers like building on 200mm fab lines because the manufacturing technologies are extremely mature and the costs are low. Many customers don’t get much benefit from moving to lower geometries, and they want to stick with the designs they’ve already paid for. Many IoT sensors and similar products are therefore built at relatively large process nodes. As demand for these products has grown, 200mm capacity has gotten difficult to book. Large foundries like TSMC have been slow to add new 200mm capacity, and 200mm utilization was already high at many fabs before the pandemic hit.
This old graph describing how *450mm* wafers were going to ramp also shows what analysts expected to happen to 200mm shipments. After 2015, 200mm demand began growing again, while 450mm wafers never launched.
I don’t want to give the impression that literally every foundry with 200mm capacity is running all-out, but it takes a significant amount of time and money to port designs to a new fab. A company with a chip designed for Foundry A’s manufacturing line would likely face significant costs to port the same chip to Foundry B. In some cases, a company would pay more to port a chip to a new foundry/node than it would ever make back from selling it. Not literally every 200mm fab is full, but the pure-play foundries with significant amounts of 200mm capacity aren’t having much trouble filling it.
The pandemic has been brutally bad for many people in the US and across the world, but the semiconductor industry has been one of the only bright spots. That translates to additional demand for chips of all sorts, which increased the pressure on a supply chain that was already running as fast as it could.
“It’s an interesting situation this year. Not only is 200mm in demand, but there is also demand in general for trailing-edge and specialty products. It’s power, CMOS image sensors, RF, and those kind of products,” Risto Puhakka, president of VLSI Research, told SemiEngineering . “There is also a division in the marketplace. If you are IDMs like TI, NXP, or the standard analog guys, 2020 has been tough. The industrial, automotive, and power markets have been in a tough place that’s been impacted by the coronavirus. But at the same time, foundries are booming. The foundries are addressing what I would call the consumer markets.”
Demand for Wi-Fi and Bluetooth chips has skyrocketed, leading to shortages of both. Demand for automotive electronics has increased more quickly than expected. Huawei soaked up a tremendous amount of manufacturing capacity earlier this year before it was cut off in early September, and manfucturers like Xiaomi tried to take advantage of the company’s weakness with increased component orders of its own. In the US, demand for PCs has been much higher.
This may be part of the answer to why it’s so hard to get a lot of equipment right now. It’s not just because COVID-19 scrambled supply, it’s because it simultaneously screwed up demand. Combine that with the need for new laptops to handle remote learning, new console launches, a much-anticipated GPU launch or two, and COVID-19, and Huawei, and now a general 200mm shortage, and it’s easier to see why the global market doesn’t seem to be running very smoothly. Another 220,000 wpm (wafers per month) worth of 200mm capacity is expected to come online in 2021, with total capacity of 6.4M wpm across the planet.
Now Read:
- Intel Tops Semiconductor Revenue, as TSMC, Nvidia, and AMD Vault Upwards
- TSMC Will Open $3.5 Billion Semiconductor Fab in Arizona
- Samsung Now Producing 16Gb LPDDR5 With Extreme Ultraviolet Lithography
No comments:
Post a Comment